The Equality Multiplier

Transcription

1 The Equality Multiplier Erling Barth and Karl O. Moene June 2008, work in progress Abstract Welfare spending and wage coordination both generate equality. Considering the welfare state as a social insurance device, we show how coordinated wage compression generates support for higher generosity. Considering wage coordination as structured negotiations, we show how welfare generosity fuels wage compression. Together the two mechanisms enforce each other generating an equality multiplier. Using data on OECD countries over the period we identify a sizeable magnitude of this multiplier. This complementarity of institutions may help explain the diversity across otherwise similar countries in the OECD area. It may also explain why countries cluster around different worlds of welfare capitalism - the Scandinavian model, the Anglo-Saxon model and the Continental model. 1 Introduction With only one third of pre-tax wage inequality of the US, the Scandinavian countries of Denmark, Norway and Sweden have twice as generous welfare spending as the US. This is a stark illustration of a general pattern across rich countries. The generosity of welfare spending is negatively associated with the inequality of wages before taxes and transfers as the scatter plot in Figure 1 shows. The negative association resembles what Peter Lindert (2004) calls the Robin Hood paradox in which redistribution from the rich to the poor is least present where it is the most needed. 1 As Figure 1 shows, countries with more wage equality tend to have more generous welfare states, while countries with more generous welfare states tend to have more wage equality. These are not words for the same thing, we claim, but rather two separate mechanisms with distinct causal effects. One mechanism the equality magnifying effect runs from the wage distribution to the determination of welfare state policies as more wage equality makes the welfare state more generous. The other mechanism the wage equalizing effect runs from welfare state policies to wage determination as more generous welfare benefits reduce wage inequality. Barth: Institute for Social Research, Oslo and ESOP, University of Oslo, erling.barth@socialresearch.no. Moene: Department of Economics, University of Oslo, k.o.moene@econ.uio.no. This work part of a larger research project at ESOP, the Frisch center and ISF. We are grateful for... Research Grant from the Research Council of Norway is gratefully acknowledged. 1 Lindert draws attention to a more general regularity than we do: Poverty policy within any one polity or jurisdiction is supposed to aid the poor more, the lower the average income and the greater the income inequality. Yet over time and space, the pattern is usually the opposite. (Lindert 2004, p 15.) This equality-generosity puzzle runs counter to the most prominent theories of welfare spending such as the seminal papers by Romer (1975), Roberts (1977), and Meltzer and Richard (1981) which all predict that higher pre-tax inequality should be associated with a more generous welfare state. 1

2 Generosity Index Welfare Generosity and Wage Inequality Figure 1: Welfare Generosity and Wage Inequality Sweden Norway Nordic Denmark Netherland Belgium Finland Italy Germany Austria France Cont. Europe Ireland New Zealand Switzerland Anglo-sax Canada Australia United Kingdom Wage Dispersion United States Generosity Index Avg. Gen. Index Note: Wage dispersion is the relative difference between the 9th decile and the 1st decile of gross hourly wage. Source: mainly OECD, see data appendix. Overall Generosity Index is an index of welfare generosity developed by Lyle Scruggs, University of Connecticut, see data appendix. Countries included: AUS, BEL, CAN, DEN, FIN, FRA, ITA, JAP, NET, NZ, NOR, SWE, SWI, UK, and the US. N=361, Years included: The two effects are complementary and enforce each other. Neither of the two can alone explain the equality generosity pattern over time and across countries. Combined, however, the two mechanisms generate a cumulative process where a more equal wage distribution fuels welfare generosity and where a more generous welfare state fuels wage equality, stimulating more welfare generosity and so on. The cumulative process adds up to a sizeable social multiplier. 2 Our paper provides a theoretical explanation of this equality multiplier and an empirical assessment of its magnitude in OECD countries. Explaining the equality magnifying effect of how the the wage distribution determines welfare generosity, we emphasize that protection against risks has been more universally sought and has been more important for the expansion of the welfare state, than pure redistribution of resources (Baldwin 1990, Barr 1992). Building on Moene and Wallerstein (2001, 2003) we focus on welfare spending as social insurance against loss of income due to sickness, unemployment, and old age. How more equality alters the political support for social insurance, depends to a large extent on how wage earners below the mean react to higher incomes. A mean preserving reduction of inequality in a skewed wage distribution, implies that the majority of voters below the mean become richer. Voters who obtain higher incomes, but who are exposed to the same risks, simply want to be better insured. The extent to which voters achieve what they want depends on the political competition between ideological parties. If a party trades off its ideological preferences against 2 Glaeser, Sacerdote and Schenkman (2003) discuss social multipliers where individual behavior depends on aggregate behaviors. In our case the complementarity is between institutions of the labor market and the welfare state. 2

3 higher odds of winning the election, it will commit to a program that compromises the interests of decisive voters. It matters which party wins the election, but all parties run on a program that is already adjusted to the compressed wage distribution. How generous the welfare state becomes, therefore depends on the outcome of the political competition over voters support, but the interests of voters are shaped by the pre-tax distribution of wages. Contingent on which party wins, it follows that a compressed wage distribution raises the generosity of the welfare state. Explaining the wage equalizing effect of welfare spending on the wage distribution, we focus on how welfare benefits strengthen weak groups in the labor market. Weak groups simply become stronger in wage negotiations as their fall back position is improved. They are therefore able to command a higher pay and to improve their relative wage. We incorporate this in a well structured bargaining framework that allows for both decentralized and more centralized wage setting. Variation across countries and over time is caused by different institutions in the labor market and different political color of governments in addition to factors that are specific to each country all of which we use in the empirical identification of the effects. We therefore need to combine the equality magnifying effect and the wage equalizing effect with these basic factors that have a direct impact on relative wages and welfare spending. The impact of wage coordination is particularly important in our story. In all countries unions have a stronger impact on the distribution of relative wages than on the distribution of profits and total wages (Freeman and Medoff 1984, Wallerstein 1999, and Moene and Wallerstein 1997). Coordination in wage setting changes the logic of wage negotiations as relative wages are to a large extent taken out of union-employer bargaining and placed into union-union bargaining. Since unions adhere to fairness norms this change strengthens the bargaining position of weak groups in the labor market. The level of wage coordination determines the units over which such fairness norms are applied. When wages are determined at the firm level, unions affect the distribution of wages within the firm. When wages are set at the industry level, unions affect the distribution of wages across firms within the industry. When wages are set at the national level, unions affect the distribution of wages across firms, industries and occupations throughout the entire nation. The total impact of different levels of wage coordination is strongly affected by the equality multiplier, which of course can go either way, depending on initial stimuli. To to assess the magnitude of the equality multiplier, and to test our propositions we use panel data of OECD countries from the period of 1976 to Since the causality between wage inequality and welfare generosity runs both ways the major empirical challenge is to identify the parameters of two equations: the equality magnifying curve, where welfare spending depends negatively on wage inequality, and a wage equalizing curve, where wage dispersion depends negatively on the generosity of the welfare state. Identification of these parameters is obtained using fixed-country effects and instrumental variable methods. Our theoretical arguments suggest that we may instrument wage inequality by utilizing measures of bargaining coordination, and that we may instrument welfare generosity by utilizing measures of right wing versus left wing orientation of the government. There are two key identifying assumptions. The first is that bargaining coordination does not affect welfare generosity for a given wage inequality and political orientation of the government. The second is that the political orientation of the government does not affect wage inequality for a given welfare generosity and wage bargaining system. 3

4 Our empirical findings support our main propositions. Societies with more equal wages before taxes and transfers tend to have more generous social insurance, whereas societies with more unequal wages tend to have less generous social insurance. As our theory suggests, institutions and mechanisms that enhances equality in the market place, also strengthens the demand for social insurance; whereas generous social insurance tend to reduce inequality in the market place. Together the two mechanisms generate an equality multiplier that magnifies external shocks by 65 percent in the long run. Accounting for such institutional complementarities and social spill-overs is important for the understanding of why some countries are more egalitarian than others. Certain policies, institutions and behaviors simply fit together and strengthen each other. In the long run, the outcomes may look as if societal arrangements come in packages with different social and economic organization. Esping Andersen s (1990) emphasis of the three worlds of welfare capitalism is one important example of such clustering that we explore further. While Esping Andersen emphasizes welfare state arrangements as more important than wage setting institutions for the clustering of countries in the three worlds, we find that wage setting institutions play a more essential role. In this respect our paper complements the study of how labor market arrangements affect the support for the welfare state by Iversen and Soskice (2001). They focus on how irreversible investments in skills with low degrees of transferability fuel the demand for social insurance against the possible future loss of income. Our paper also connects to the more general discussion of the economic determinants of welfare spending and inequality (see for instance Cameroon 1978, and Katzenstein 1985 for the role of openness and country size). Our paper complements the more general discussion in Lindert (2004) of why the rise of welfare spending was so much higher in some countries than in others, and of why not all countries have a European style welfare state (Alesina and Glaeser 2001, 2004). In section 2 we provide a simple exposition of the equality magnifying effect of wage compression on welfare spending, while we in section 3 discuss the wage equalizing effect of welfare spending on the wage distribution. The simple set-up of Section 3 also provides the key explanation why wage coordination leads to wage compression and thus why some countries compress their wage structure more than others. Section 4 derives the equality multiplier and section 5 provides data discussion, the empirical assessments of the multiplier and the tests of our predictions. 2 The Equality Magnifying Effect The welfare state is not the machinery for redistribution that many thinks (Romer 1975, Roberts 1977, Meltzer and Richard Pure redistribution is seldom accepted as legitimate. Policies that, in addition to providing a more fair distribution, cover social demands for which the market fails to provide, are much more likely to be both legitimate and popular. The emergence of the modern welfare state seem to confirm this assertion. Assumption 1. Protection against risk is more important as a justification for welfare spending than pure redistribution. Here we take this view and focus on how the welfare state provides social insurance against loss of income due to sickness, unemployment, and old age. We also restrict the discussion to another stylized fact: Assumption 2. Rich voters prefer lower welfare spending than the poor. 4

5 Numerous opinion surveys in OECD countries show that high-paid wage and salary earners, prefer lower taxes (and lower welfare benefits) than low-paid wage earners. The reason is most likely that the high-paid are less exposed to risks of income loss than the low-paid, and therefore have less interests of a generous welfare state. With these two assumptions we consider a society with a number of citizens normalized to 1. We order their earnings according to their position in the wage distribution denoted i. There are m positions and they are numbered such that w 1 w 2...w i... w m. The share of the work force in position i is n i. Letting (1 e i ) indicate the risk of loosing ones income in position i, we define ē = m n i e i (1) i=1 as the average fraction of citizens who are working, where e i denotes the fraction of workers in position i that are employed. The average wage is w = m n i w i (2) i=1 We denote by g the generosity of the welfare system. In all welfare systems social insurance is offered on better terms for low wage earners than for high wage earners. We incorporate this by assuming that each worker who loses his income obtains welfare benefits equal to g. This is of course a grave simplification, but one that can easily be modified. 3 The welfare benefits are financed by a constant marginal tax t on total national income (wages plus profits), e i n i p i = ẽp, where p i is the productivity of workers in position i, where P = n i p i is the average productivity in the the economy, and where ẽ is the productivity adjusted employment rate ẽ > ē (since e i is positively correlated with p i ). We think of ẽp as representing gdp per capita. The balanced budget equation is (1 ē) tẽp = (1 ē)g = t = g γg (3) ẽp Each citizen has an utility function with a constant relative risk aversion µ over consumption U (c) 1 1 µ c1 µ with µ > 1 Since the individual risk of income loss must be considered a serious threat to the living conditions of a typical voter, we limit the discussion to cases where citizens have a 3 In general, some benefits are proportional to present earnings or past contributions; others are not. We could have incorporated this by a given parameter θ (0, 1] reflecting the composition of welfare spending and the extent to which the poor are offered social insurance on better terms than the rich: ( g i = θ + (1 θ) w i w The benefits g (the benefit level to workers with the average wage) of the social insurance scheme are distributed with a fixed component common to all and a variable component that depends on past and present contributions. The fixed component is θg which defines the floor of welfare benefits to people without income. The variable component is proportional to income relative to the mean g (1 θ) w i / w, implying that here g i is the welfare benefits to a worker in position i in the event of income loss. The higher is θ the more redistributive is the terms of the social insurance scheme. In the presentation we apply the simplifying assumption that θ = 1. ) g 5

6 relatively high degree of risk aversion µ > 1. Inserting c i = (1 t)w i and t = γg the preferences of a worker with a wage w i and risk (1 e i ) is v i (g) = e i U ((1 γg) w i ) + (1 e i )U (g) (4) We find worker i s preferred generosity of the welfare state - his ideal policy - from the first order condition v i / g = 0 which after some rearranging (and utilizing the constant elasticity of the utility function) can be written as [ ] 1 1 ei µ 1 wi x i P = e i [ 1 ē ẽ ] 1 [ ] µ µ 1 g µ 1 i h (gi ) (5) P (1 γg i ) where g i is the ideal policy of voters in position i. In (5) x i depends on the gross income of the voter, relative to the average income P, and his odds (1 e i )/e i of loosing the income. We denote it the odds corrected relative pay. Given the simultaneous distribution of individual wages and risks over the population, we can in principle derive the distribution of the ratio x i over the population. Assumption 2 above, stating that richer voters prefer lower taxes, implies that x i is declining in i. The voter with the median value x i = x M prefers a value of g that we denote g M, where x M = h (g M ). When wage compression raises the wages below the mean, higher values of w i for given risks e i imply that the value of the odds corrected relative pay x M goes up. Since h (g) > 0, this implies that the preferred level of g goes up as well. Hence, with a skewed wage distribution, where the median wage is below the mean, a mean preserving compression of wage differentials implies a higher ideal point of welfare spending for a majority of the voters. Note that wage compression means higher wages for positions below the mean holding the risks of these positions constant. The main intuition why wage compressing increases the demand for welfare spending, is therefore that the majority of voters, the high risk workers, demand better social insurance. Note also that each level of g i is increasing in the average income of the country. 2.1 Political competition With two parties or blocks - left and right - that differ in their ideologies, median voter results are not directly applicable. We follow the approach of Whittman (1977) as discussed in Roemer (2001). The ideology of parties may be based on the interests of the parties core groups. We write these preferences over policy outcomes as v L (g) and v R (g) which are maximized for g = G L for the left party and g = G R for the right. It is natural to assume that the two parties ideologies are located on each side of the ideal point of the median voter: G R < g M < G L (6) Each party is willing to compromise somewhat on ideology in order to improve the chances of winning the election. In the language of John Roemer each party is reformist in the sense that it aims at maximizing the expected party utility (V L, V R ). Letting the probability that the left wins when it proposes g L and the right proposes g R, be denoted p = p (g L, g R ) (to be derived), we have V L = pv L (g L ) + [1 p] v L (g R ) (7) V R = [1 p] v R (g R ) + pv R (g L ) (8) 6

7 In our framework unions (and employer associations) do not play a distinct role in the determination of welfare spending. This is not because we think that these interest organizations are not lobbying for welfare policies. They may very well be active, but in a way that obeys their members welfare state preferences. This is part of the reason why political parties are likely to have a realistic assessment of their vote shares for different policies. To find the winning probabilities we need to know who votes for which party for all relevant proposals g L and g R. If each party should propose the median policy g = g M, both would obtain an expected vote share equal to 1/2. If g L > g R voters with an interest to vote left must have v i (g L ) > v i (g R ) Using again the constant elasticity of the utility function this inequality is equivalent to [ ] 1 U (P (1 γgr )) U (P (1 γg L )) µ 1 x i > k (gl, g R ) for g L g R (9) U (g L ) U (g R ) where x i is the odds corrected relative pay as defined by (5). Here k (g L, g R ) is the threshold for the left right votes on welfare policies. This threshold depends on the income of the average income earner. The numerator inside the brackets is his utility gain of having the right party s versus the left party s tax proposal when working. The de-numerator is his utility gain of having the left party s versus the right party s benefit proposal when not working. A large tax gain would lead to a low support for the left party, while a large benefit gain would lead to a high support for the left party. The threshold k (g L, g R ) is increasing in g L and declining in g R (see appendix). Denote by F the cumulative distribution of x i in the population. We then have that the expected vote shares are { 1/2 when g s L = 1 s R = L = g R (10) 1 F (k (g L, g R )) when g L g R The actual vote shares, however, may be affected by random events that we capture by applying a variant of probabilistic voting (see Roemer, 2001). The actual vote share of the left is equal to the expected vote shares s L plus a stochastic error term ε and the actual vote share of the right is the expected vote share s R minus the stochastic term ε. The value of ε has a symmetric distribution around Eε = 0. The actual vote shares are s L + ε and s R ε. The probability that the left wins is thus p = p (g L, g R ) = Pr (s L + ε > 1/2) (11) and the probability that the right wins is (1 p). In the political competition each party is willing to deviate somewhat from its ideal policies as long as the chances of winning go sufficiently up. The trade-offs involved are captured by the first order conditions p g L [v L (gl) v L (gr)] + p v L g L = 0 (12) p g R [v R (g R ) v R (g L )] + (1 p) v R g R = 0 (13) Together the two equations define the Nash equilibrium of the policy game. The ideal policy of the median voter is not an equilibrium outcome since, for g R = g M it pays for 7

8 the left party to deviate from g M by setting a higher level of g. This is so since the marginal ideological gain p v L / g L is strictly positive. By increasing the level of g L the left party s chance of winning the election declines and p/ g L < 0. In optimum (with the right guesses on the opponent s policy choice g R ) the left party chooses g L = g L > g M such that the marginal reduction in the chance of winning the election times the gain of winning [v L (gl) v L (gr)], just equals the marginal ideological gain of running with a policy closer to the party s ideals. If the ideal policy of the median voter g M reflects a higher level of generosity, due to wage compression, the left party would as a response change its policy in the same direction. Similarly the right party would deviate from g R = g M by reducing the level of g R in the direction of the party s ideal policy. In optimum (with the right guesses on the opponent s policy choice g L ) the right party chooses g R = g R < g M such that the marginal reduction in the chance of winning the election times the gain of winning [v R (g R ) v R (g L )], equals the marginal ideological gain of running with a policy closer to the party s ideals. The discussion so far is summed up in the following proposition: Proposition 1 Wage compression induces a higher level of welfare generosity: The outcome of the political competition is g L > g M > g R, and wage compression induces a new equilibrium (a higher level of g M ) with higher levels of both g L and g R. This is our theoretical account of the equality magnifying effect, stating that more equality in the labor market magnifies the generosity of welfare spending. Wage compression implies that voters become more similar in their demands towards the welfare state and the majority of them want a higher level of social insurance. Political parties therefore offer more generous welfare policies to compete for voter support in elections. The implemented generosity of welfare spending depends on which party wins the election. But both blocks have adjusted their policies to the more egalitarian wage distribution implying higher welfare spending contingent upon party in power. 3 The Wage Equalization Effect Welfare benefits impose an implicit minimum wage ω (g) in the labor market. This implicit minimum wage can be derived from the participation constraint. Assume that the basic preferences are over income and leisure. Let bu(c) indicate the utility of net income c, where b > 1 when not working, and b = 1 when working. 4 The participation constraint is then U ((1 γg) w) bu (g) = w ω (g) g 1 γg b 1 1 µ (14) where ω (g) > 0. Thus the generosity of the welfare state establishes a floor which eliminates the lowest wages, and the floor is higher the more generous the welfare state. Generous welfare benefits have this wage equalization effect both in coordinated and non-coordinated wage systems. We incorporate the effect within a set-up with heterogenous job productivity. The productivity per worker in position i is denoted p i. To assess the equality multiplier we must derive how both coordination and welfare spending affect the wage distribution. 4 The inclusion of b would change nothing in section 2 above. 8

9 3.1 Decentralized wages Non-coordinated or decentralized wage setting is considered as rent sharing where workers share is denoted α. Hence, in position i each worker obtains w i = max [αp i, ω (g)] (15) The parameter α can be thought about as the bargaining power of union locals. Decentralized wage setting, however, can result in unequal pay for equal work even without unions as the cost of filling vacancies or of training new workers may endow workers with bargaining power. Empirical work on relative wages in the US, for instance, reveals large interfirm and inter industry wage differentials that cannot be explained by union membership or any other observable characteristics of the job or the workers (Krueger and Summers 1988, Groshen 1991, Gibbons and Katz 1992). We number the jobs such that p 1 p 2... p m, and denote by q the position with the highest productivity that is paying the implicit minimum wage ω (g) implying that w 1 = w 2 =... = w q = ω (g) (16) Obviously, a higher implicit minimum wage, due to higher welfare generosity g, compresses the wage scale. Average productivity, defined as above, P = n i p i is assumed to be the same across bargaining regimes. Since we basically are concerned with relative wages this is less restrictive than it may seem at first glance. We define π i = p i w i, and the aggregates are denoted π = n i π i and w = n i w i where π + w = P. Clearly, a higher welfare generosity g raises the average wage w and lowers average profits π. 3.2 Coordinated Wages Coordination implies that wages to some extent are taken out of local competition and placed into a system of collective decision making. This alters the structure of who negotiates with whom. The basic change is that many union employer bargains are replaced by union union arguing or bargaining. Using capital letters to indicate the outcome of coordination, wage coordination can be thought of as a system where: (i) the employers association negotiate with the union confederation about the average wage (the total wage bill) W = n i W i with bargaining power α on the union side and 1 α on the employer side. (ii) the total wage bill W is distributed between the unions via collective union-union bargaining. In the case of a break-down of wage coordination the non-cooperative benchmark works as a fall-back position. There is an expected status quo bias in the sense that there might be delays before the non-cooperative system is in place, implying that the value of the fall back positions is diminished by a factor δ < 1. Applying the generalized Nash bargaining solution we write the Nash product N = [ n i (p i W i δπ i ) i ] 1 α [(n 1 W 1 δn 1 w 1 ) β 1 (n 2 W 2 δn 2 w 2 ) β 2... (n m W m δn m w m ) βm ] α 9

10 Here the parameter β i represents union i s strength in union-union bargaining, where i β i = 1. In our highly structured wage coordination, unions have much more influence over relative wages than employers as union-union bargaining takes over for separate union-employer bargaining. Each union, in the non-coordinated case, is assumed to be equally strong towards their employers (i.e. having the same bargaining power α). This does not imply that each union, in union-union bargaining, is equally strong towards other unions. In union-union bargaining the effective strength β i of each union must be legitimate - based on acceptable principles that can be defended publicly. Below we make clear how these effective strengths reflect bargaining power moderated by fairness norms. Maximizing N with respect to W i, we obtain (1 α) (n i W i δn i w i ) = αβ i n i (p i W i δπ i ) β i α (P W δπ) i Summing over i and utilizing that π + w = P we obtain the following expression of the total wage bill W = δ w + α (1 δ) P (17) To interpret this, recall that the union confederation can guarantee itself the fall back pay-off δ w - the first term in the expression. The second term stems from the potential loss of δp associated with a possible break down of coordination as the unions obtain their share α of the gain of no breakdown (1 δ)p. Equation (17) can also be written as W = w + (1 δ) (αp w) which shows that W w since αp w. Thus in our set-up wage coordination is associated with wage moderation whenever the implicit minimum wage ω(g) is binding for at least one group in the non-coordinated case. In other words the generosity of the welfare state increases both w and W, but the rise in the coordinated average wage W is less than the rise in the non-coordinated average wage w. The first order conditions for max N also imply β i n k (W k δw k ) = β k n i (W i δw i ). Summing over k and dividing by n i we obtain the wage in position i W i = δw i + β i n i α (1 δ) P (18) where β i /n i is the strength per worker of union i. The equation says that the wage level to group i with coordinated wage setting is the value of the fall back position plus a share β i /n i of the total gain to all unions of not letting wage coordination break down. Notice from (18) and (15) that W i δαp i + β i n i α (1 δ) P. The wage level to group i with coordinated wage setting is thus greater or equal to a weighted average of the local productivity of the group p i and the total productivity of all groups P. The inequality is strict whenever the outside option is binding in the non-coordinated case. 3.3 Fairness The effective strength of each union β i is influenced by conceptions of fairness such as equity or equal treatment, equal wages for equal work, rewards according to productivity. These social norms concern relations between workers in different positions. Unions may always apply such fairness norms, but only over their relevant bargaining units. Coordinated wage setting makes these norms more visible and more effective as coordination extends the unit over which they are applied. This is why fairness norms seem to be less important in local union-employer bargaining than in coordinated bargaining. 10

11 Here the values of β i are assumed to be a compromise between two fairness principles: (i) similar treatment, implying that the strength per worker tends to be equal across unions, and (ii) reward according to productivity, implying that the strength per worker tends to reflect the contribution measured by the local p i relative to average productivity. 5 We capture the trade-off between these two principles by assuming that the effective strength in union union bargaining is an weighted average of union size and relative local productivity: p i β i = rn i + (1 r) n i with 0 r 1 (19) P r = 1 implies that all weight is placed on the concern for similar treatment and each union is equally strong per member, whereas r = 0 implies that all weight is placed on contribution as reflected in local productivity. Whenever some weight is put on similar treatment r > 0, the wage scale is compressed by coordination. To show this we first consider positions where the implicit minimum wage is not binding. Then W i = w i + (1 δ)rα[p p i ] (20) Observe that if the strength of each union is determined only by its local productivity, that is r = 0, wage coordination just reproduces the non-coordinated wage structure. With some weight on similar treatment, however, wage coordination implies that jobs with productivity less than the average, p i < P, obtain a wage rise, while jobs with productivity above average, p i > P, obtain a wage decline. Hence, wage differentials are compressed both from below and above. Consider next positions where the implicit minimum wage is binding in the noncoordinated case, i.e. jobs where αp i < ω(g). In these jobs the coordinated wage is W i = ω(g) + (1 δ) max [0, rα(p p i ) + αp i ω(g)] (21) Hence, W i ω(g) = w i. Wage coordination leaves the low wage unchanged at its minimum level ω(g), if the weight r on similar treatment is sufficiently low. If the weight on similar treatment is sufficiently high, however, wage coordination implies a rise to W i = δω(g) + (1 δ)αp i + (1 δ)rα(p p i ). The relative wage between a high productivity job i = H and a (sufficiently) low productivity job i = L, can be written as I HL = W H W L = w H + (1 δ)rα[p p H ] ω(g) + (1 δ) max [0, rα(p p i ) + αp i ω(g)] (22) This expression can be thought of as a measure of the inequality between two fractals in the wage distribution. Clearly, wage coordination reduces this inequality by lowering the high wage, and in addition either raising the low wage or leaving it unchanged. By putting it somewhat differently, union strength in union-union bargaining, β i, may reflect both political bargaining power (the size of the union n i )) and economic force (the productivity of it s members p i ). The weight r put on political bargaining power, obviously benefit low wage workers. The value of r is likely to be strictly positive since all groups - also the lowest paid - can inflict a cost on the others by not cooperating. The value of r is likely to be strictly less than 1 since economic force easily translates into sharing power. 5 The two principles can be denoted equity and meritocracy. 11

12 From the expression of I HL we also see that a higher welfare generosity lowers wage inequality, as low wages before taxes and transfers go up with the generosity g, and high wages remain unchanged. Higher generosity improves the outside option of the workers and therefore leads to a higher implicit minimum wage. This minimum wage can be binding both in the non-coordinated and the coordinated wage structure. If the minimum wage binds in the coordinated case, higher generosity raises low wages directly. If the implicit minimum wage is only binding in the decentralized case, higher generosity improves the bargaining outcome of low paid workers and raise their wages beyond the implicit minimum wage due to an improved fall back position in union-union bargaining. This discussion is summed up in the following proposition: Proposition 2 (i) The inequality between high and low wages declines with welfare generosity irrespective of the level of coordination in wage setting and the weight r put on similar treatment in the effective union strength. (ii) Wage coordination compresses wage differentials between high and low wages: workers in jobs with above average productivity (p i > P )obtain lower wages, while workers in jobs with productivity below the average (p i < P ) obtain higher wages relative to the non-cooperative benchmark. This is our theoretical account of the wage equalizing effect. Part (i) of the proposition shows how more generous welfare benefits lead to wage compression from below. Part (ii) demonstrates how wage coordination leads to wage compression from below and above. So far we have focused on extreme cases with and without wage coordination. Intermediate levels of partial coordination are also highly relevant. The implication of intermediate levels of wage coordination can easily be derived from our set-up. We have shown that wage coordination compresses wage differentials over the bargaining unit. If the bargaining unit does not include all positions, the excluded positions are likely to be remunerated by local systems or sharing rules. This is particularly relevant for some high paid non-union positions, implying that the wage distribution becomes skewed with a thin right tale - and with a median below the mean wage. More generally, compared to the decentralized system, intermediate levels of wage coordination unambiguously lead to lower wage inequality whenever the coordination includes positions below and above the mean. As the fairness norms held by unions become more visible and pronounced the more coordinated the wage setting system, the value of r can depend on the level of coordination. In highly coordinated wage systems union representatives must publicly defend the relative wages they have negotiated. Thus the pressure on equal treatment may become more severe. If this is so, there is an additional channel that strengthen the impact of coordination on wage compression. Finally, the combination of welfare spending and wage coordination implies overall wage moderation whenever the implicit minimum wage imposed by the welfare state is binding for some groups. While, in the non-coordinated case all costs associated with higher implicit minimum wages are borne by employers, more of the costs are borne by the workers with wage coordination. 4 The (In)Equality Multiplier 12

13 The equality magnifying effect, discussed in Section 2, shows how the level of political support for a generous welfare state is raised by the degree of wage equality. In short the mechanism can be written as g = g (I; z) where I is wage inequality, and z is a vector of exogenous factors. As proposition 1 shows, higher wage inequality lowers the generosity of the welfare state, i.e. g (I; z) < 0. Our model also suggests that the ideological orientation of the winning party should be included in the vector z in addition to the level of gdp per capita and indicators that pick up the economic risks that voters are exposed to - such as economic openness. The wage equalizing effect, discussed in Section 3, shows the reverse relationship between inequality and the generosity of the welfare state. In short this mechanism can be written as I = f (g; y) where y is a vector of exogenous factors. As proposition 2 shows, higher welfare benefits reduce wage inequality, i.e. f (g; y) < 0. Our model also suggest that the level of coordination in wage-setting should be included among the vector x in addition to the level of gdp per capita and the level of union density indicating the number of positions included in wage coordination. Combining the two mechanisms an institutional equilibrium (g, I ) can be obtained: For a given level of welfare generosity g, the wage inequality I is generated; for a given wage inequality I the welfare generosity g is generated. This equilibrium can be reached by gradual adjustments - a cumulative sequence of wage settlements and welfare state adjustments. The mechanisms contain an equality or inequality multiplier depending on stimuli between wage setting and welfare spending. The multiplier summarizes the feed back mechanisms between the equality magnifying effect and the wage equalizing effect. Any shifts in one of the two curves is then magnified by the multiplier [ ] 1 m = (23) 1 f g where m is greater than one whenever the system is stable, i.e. f g < 1. We now set out to estimate the multiplier. 5 Evidence 5.1 Identification Since the causality between wage inequality I and welfare generosity g runs both ways the major empirical challenge is to identify the parameters of the two equations: I = f(g; y) and g = g(i; z) (24) We need some exogenous factors that are included in y and thus affect wage inequality, but do not affect welfare generosity; and some exogenous factors that are included in z and thus affect welfare generosity, but do not affect wage inequality. Our theoretical model suggests that the political color of the government should affect welfare generosity, but not wage inequality, and that the level of wage coordination should affect wage inequality, but not welfare generosity. We use these restrictions to identify the the slopes of the two curves. Our main identifying assumptions are that conditional on the other explanatory variables and country fixed effects, the impact of unions and employer associations on welfare generosity does not depend on the bargaining system, 13

14 and similarly that the impact of the government on wage inequality is not through direct intervention. There are examples that seemingly go against the assumption that government does not affect wage inequality. The Thatcher government, for instance, clearly affected wage inequality in the UK. The way it did this, however, does not contradict our assumptions. The government affected wage inequality by changing the regulations of how unions could operate and how they could recruit members. The effect on wage inequality is therefore indirect through changes in bargaining system and in union density, variables that we do include in the vector y. Another example is the recent policy changes in Sweden where the right wing government is effectively dismantling the so called Ghent system of unemployment compensation. In the Ghent system unions administer funds for unemployment insurance that are subsidized by the government. Several studies show that union density is higher in countries with the Ghent system (Lesh 2004, Holmlund and Lundberg 1999, and Bøckerman and Uusitalo, 2005). Again the way the government affects wage inequality - recently first in Finland in the 1990s and maybe now in Sweden - does not contradict our assumptions. The possible effects on wage inequality go indirectly through changes in union density, which again is included in the vector y. There are also examples that seemingly go against the assumption that unions and employer associations do not affect welfare spending. There are lobbying efforts for specific welfare state policies both from union confederations and employer associations. Comprehensive unions are for instance sometimes seen as strong defenders of the welfare state. Their impact on welfare policies, however, are strongest when they lobby for the interests of the majority of the electorate. When they lobby for more special interests, the problem is credible threats and credible promises. Unions and employer associations cannot guarantee reelection of politicians and governments as they have no control over how votes are cast. This limits their ability to influence social policies systematically against the interests of the electorate. Low density unions have no impact, whereas high density unions within a comprehensive union movement have their own political conflicts over welfare spending that mirror the political conflicts of voters in the electorate. In general, there is a long European tradition, in particular in the Nordic countries, that the government does not intervene in wage setting which is left to be negotiated by the organizations of the labor market. The principle of non-intervention endorsed by the organized interests in the labor market also limits their ability to affect social policies directly. These identifying assumptions are used to form instrumental variables in our empirical analysis. Details on the empirical strategy is provided below. 5.2 Data We use a panel of observations from 18 OECD countries to test key predictions from the model and to quantify the size of the equality multiplier. Our main results are obtained using 359 observations of yearx country cells from the period of Detailed descriptions of the data are provided in the appendix. Wage inequality is measured by the relative difference between the 9th decile and the 1st decile of gross hourly earnings. This measure is gross of taxes and transfers, and based on individual outcomes in the labor market. Most of the observations of wage inequality 14

15 are obtained from OECD (Employment Outlook and Society at a Glance), which reports d9d1 ratios. Wage inequality data are supplemented with observations from two other sources. Details are provided in the appendix. Table 5 provides a description of the ratio between the 9th decile (d9) and the first decile (d1) of pre-tax wages of the OECD countries from 1975 to Figure 2: Wage Inequality Country Dataset Australia OECD Austria OECD ECHP Belgium OECD ECHP Canada OECD Chzechia OECD Denmark OECD NOS-S ECHP Finland OECD NOS-S ECHP France OECD ECHP Germany OECD ECHP 2.88 Greece OECD 1.80 ECHP Hungary OECD Iceland NOS-S Ireland OECD ECHP Italy OECD ECHP Japan OECD Korea OECD Netherlands OECD ECHP New Zealand OECD Norway OECD NOS-S Poland OECD Portugal OECD ECHP Sweden OECD NOS-S Spain OECD ECHP Switzerland OECD United Kingdom OECD ECHP 2.88 United States OECD OECD average OECD Note: Five years averages of available data. Each cell does not necessarily represent data from each of the five years in the interval. OECD average is an average of the numbers obtained from OECD sources. See data section for details. We find large differences in pre-tax wage inequality across countries. Using Esping Andersen s (1990) country classification 6 of Three Worlds of Welfare Capitalism, we find an average level of 3.4 for the Liberal countries, 2.9 for the Conservative countries, while the Social Democratic group of countries have an average of 2.3 in our data. Out of 6 See data section for the classification of countries. 15

16 the 20 countries, 7 have experienced an increase in wage inequality from the first to the last of the observed 5-year periods, 4 have experienced stability, while 9 countries have experienced a decline in wage inequality. Generosity of the welfare state is measured by the overall generosity index provided in the Comparative Welfare Entitlements Dataset, constructed and generously made available for other researchers by Lyle Scruggs at the University of Connecticut. The index captures the generosity of income support in the case of illness, of unemployment and of disability pensions (including old age) of each country year cell. Generosity is constructed using both the replacement ratio, coverage, entitlements and timing of different schemes. As robustness tests, we also provide some figures using alternative measures of the two key outcomes. Figure 2 displays the trend in the overall generosity index as well as in public social welfare spending for each country in our sample. Again we find considerable differences across countries. Averaging the overall generosity index across the country groups gives 21.3 for the Liberal countries, 25.9 for the conservative countries and 36.2 for the Social Democratic countries. The figure also shows the trend in public social expenditures as reported by the OECD. Public spending is a measure of the outlays associated with any given level of generosity, while the overall generosity index measures the generosity of the system, as reflected in the rules concerning replacement rates, coverage, entitlements, and timing. While spending varies with economic conditions, such as the business cycle, the generosity index varies only as the rules of the system changes. We find, for instance, that both Sweden and Finland experienced a dramatic growth in public spending associated with the economic downturn of the early 1990 s in these two countries, while at the same time the generosity index is on a slow decline, reflecting a tightening of the rules of the welfare system. Key variables to provide independent variation in welfare spending are indicators of right versus left wing power in government, obtained from E. Huber et. al. (2004), Comparative Welfare States Dataset and from Armingeon et. al. (2007) Comparative Political Data Set. Key variables to provide independent variation in wage inequality are indicators of bargaining systems such as union density and bargaining coordination, obtained from the Golden, Miriam; Peter Lange; and Michael Wallerstein data set(see Golden et al, 2006). We also use workers involved in labor disputes (strikes and lockouts) as an indicator of less (or failed) wage coordination. These data are taken from Armingeon et al (2007). 7 Remaining explanatory variables, such as openness, GDP per capita, the share of the population with tertiary education and the employment rate of the population are detailed in the appendix as well. 5.3 Inequality and Generosity Figure 1 in the introduction displayed a negative bivariate association between wage inequality and the overall generosity score. This section checks the robustness of this negative association, using both various measures of inequality and generosity and a simple descriptive regression analysis. To show that the negative relationship found in figure 1 is not just an artifact of our particular measure of wage dispersion, we show in 7 Period covered: Source: until 2000: Huber et al. (2004); 2001 onwards and for Greece, Iceland, Luxembourg, Portugal and Spain entirely: ILO Laborsta Internet, (Download: ). 16